Official RunInfra SDK (TypeScript + Python) — optimized inference deployments

An agent harness that compiles a model into one provably-correct, self-retargeting CUDA megakernel and self-tunes it past cuBLAS at batch-1 LLM decode.

Open-source Agent Operating System

Memory-bounded compressed sparse attention via streaming top-k. Triton kernels for the DeepSeek-V4 lightning indexer. 32x regime extension on a single H200 | by RightNow https://www.rightnowai.co/

Dynamic weight generation for recursive transformers via input-conditioned LoRA modulation

Hierarchical Causal Latent State Machines for Object-Centric World Modeling

Autoresearch for GPU kernels. Give it any PyTorch model, go to sleep, wake up to optimized Triton kernels.

Dynamic per-token early exit for LLM inference. Skip layers tokens don't need

Pure Triton kernels for Qwen3.5-27B inference on NVIDIA B200

Run a 1-billion parameter LLM on a 10ドル board with 256MB RAM